A local-first AI research assistant featuring multi-LLM support, 20+ research strategies, multi-search-engine integration, and automated quality scoring for 212K+ academic sources, producing citation-backed PDF/Markdown reports via CLI, Web UI, REST API, or MCP Server.
Local Deep Research is a full-featured, local-first deep research platform. It offers four research modes ranging from 30-second quick summaries to full report generation, with 20+ strategies including a langgraph-agent autonomous mode where the LLM independently decides what to search and which engine to use. On the search side, it integrates free academic sources (arXiv, PubMed, Semantic Scholar, Wikipedia) and paid engines (Tavily, Google, Brave), while supporting enterprise knowledge base retrieval via LangChain Retrievers (FAISS, Chroma, Pinecone, Weaviate, Elasticsearch). Its Journal Quality System automatically scores academic sources using OpenAlex, DOAJ, and Stop Predatory Journals datasets. The backend is built on Flask + SQLAlchemy + SQLCipher (AES-256 encryption with per-user isolation), with a Vite-based frontend delivering real-time progress via WebSocket. Content extraction uses a multi-engine pipeline (crawl4ai, trafilatura, playwright, pdfplumber, etc.). Security measures include Cosign image signing, SLSA provenance, SBOM, Semgrep SAST, XXE prevention, and HTML sanitization. Deployment supports Docker Compose (with GPU override), pip installation, and a single-command full-stack Docker setup. Beyond CLI and Web UI, it exposes a REST API and an MCP Server (STDIO protocol for Claude Desktop / Claude Code), plus scheduled topic-based news digests. Engineering practices include PDM for builds, pytest + hypothesis + mypy for quality assurance.